I've chased a number of "ghost" jobs away by altering the Run Queue either it was in or of the current CPU hog - the job was getting CPU-choked and couldn't get enough CPU to complete the abort.